Cyber Attacks Detection, Prevention, and Source Localization in Digital Substation Communication using Hybrid Statistical-Deep Learning
The digital transformation of power systems is accelerating the adoption of IEC 61850 standard. However, its communication protocols lack built-in authentication and encryption, leaving them vulnerable to Man-in-the-Middle (MitM) and malicious frame injection attacks that can disrupt protection schemes operation. While most existing research focuses on detecting cyber attacks in digital substations, intrusion prevention systems have been largely overlooked due to concerns about potential network disruptions. To address this gap, this paper proposes an integrated hybrid statistical-deep learning method for detecting, preventing, and localizing IEC 61850 Sampled Values (SV)-based cyber attacks. The method models SV frames arrival times using exponentially modified Gaussian distributions and prevents malicious frames from reaching the targeted Intelligent Electronic Devices (IEDs). Malicious frames are dropped with minimal processing overhead and latency, while the method remains robust to network latency, jitter, and time-synchronization issues, and ensures a near-zero false positive rate under non-attack conditions. Long short-term memory and Elman recurrent neural networks are used to identify anomalous variations in the estimated probability distributions across IEDs for detecting and localizing MitM attacks on SV streams. The proposed method is validated across three testbeds comprising industrial-grade communication and protection devices, hardware-in-the-loop simulations, virtualized IEDs and merging units, and high-fidelity emulated networks. Results demonstrate the method's practicality and effectiveness for deployment in IEC 61850-compliant digital substations.